Dara Oseyemi
MIT Department: Media Arts and Sciences
Faculty Advisor: Prof. Hugh Herr
Research Supervisor: Christopher Shallal, Rick Casler
Undergraduate Institution: Princeton University
Hometown: Lower Merion, Pennsylvania
Biography
Dara Oseyemi majors in Electrical and Computer Engineering and has certificates (minors) in Applications of Computing, Cognitive Science, and Robotics and Intelligent Systems at Princeton University. In the summer of 2022, Dara participated in the Princeton-Intel REU program under Prof. David Wentzlaff, focusing on the performance comparison of scalar and vectorized algorithms of Matrix Operations and Polynomial Root-Finding programs for faster encryption and hashing. Last summer, Dara participated in the MIT Summer Research Program (MSRP) under Prof. Justin Solomon and worked on a project to optimize a Laplace operator for sparse point clouds. Using machine learning, she aimed to create a Laplace operator that matched the eigenvalues of the cotangent Laplace matrix for dense meshes, improving accuracy for sparse point clouds. This project deepened her understanding of machine learning applications in computational geometry. Dara is passionate about robotics and AI/ML and believes in their potential for innovation. She mentored young girls in STEM through Girls in Science & Technology (GIST) and served as an undergraduate teaching assistant and tutor at her school. Her research and mentorship experiences have equipped her with the skills, motivation, and collaborative spirit to excel in graduate research and contribute to technological advancements.
Abstract
Enhancing Prosthetic Control Through Magnetomicrometry:
Precision Tracking of Magnetic Beads in Muscles
Dara Oseyemi1, Chris Shallal2, Rick Casler3, Hugh Herr3
1Department of Electrical and Computer Engineering, Princeton University
2Department of Health Sciences and Technology, Massachusetts Institution of Technology
3Department of Media Arts and Sciences, Massachusetts Institution of Technology
Currently, human-in-the-loop prosthetics primarily use Electromyography (EMG) signals for control, but commercial leg prosthetics don’t utilize EMG due to latency, signal complexity, ec. An alternative approach involves controlling a prosthetic leg by tracking muscle lengths in the residual limb using implanted magnets. Using magnetomicrometry, we can obtain the magnets’ states from the measured magnetic fields. We need a fast and precise algorithm to accurately determine the state of the magnets from the magnetic fields. While the Levenberg–Marquardt algorithm meets these requirements, its performance deteriorates when the magnets are too distant from the sensors, leading to inconsistency or convergence failure. To address this, we employed the Extended Kalman Filter (EKF), a nonlinear version of the Kalman Filter, which estimates states over time and updates them based on previous estimates while considering thesystem’s dynamics. Our method involves designing a simulation to accurately assess magnetic beads’ position and dipole orientation from magnetometer array data. Preliminary results indicate that EKF significantly improves position estimation accuracy, even when the magnetic beads aredistant from the sensors. The implications of this research extend to the development of more intuitive and responsive prosthetic limbs, which can be potentially generalized to various types of amputations.
